advanced atm qos router setup to reduce bufferbloat

[#1]

Thanks, so I can see what it's doing now and it seems sane enough apart from not working with ATM overheads.

It is trying compensate for this by  reducing the rates the rates you asked for when setting up the shapers.

Does this setup, like on the old router, mostly work but not always?

A possible fail case on your line revolves around it being highly asymmetric. The way the shapers are set at IP rates means they don't see the traffic as it is on the line, they back off rates to compensate but it won't always be enough,

example: a bulk upload on a 900kbit upstream sync.

Sending 1492 byte IP packets actually consumes 1696 bytes at atm level. So 900k/8/1696 gives how many pps you can do roughly 66 per second, 66x8x1492 gives roughly 792kbit/s at ip level.

Small packets fare worse though, tcp acks are 40 bytes and with large downstream vs upstream can eat quite a chunk of upstream. Normally one ack for every 2 incoming - but sometimes  it's briefly 1:1 when there has been loss.

Your line (and vdsl 40/2)  if you work with full  down speed and assume 1:1 acks don't even have enough upstream bandwidth  for a download. OK so that's hypothetical and brief - but still not a nice situation to start with.

Repeating the above calculation for 40 byte acks which for ATM are 106 bytes long gives an ip rate of

900kbit/8/106 = 1061pps,  x 40 x 8 = 340kbit = a pain to choose the right number for upstream at IP level without being

miles over/under depending on packet size.

Hopefully the archer may let you work around this - but it would need entering qos rules by cli and hoping it didn't overwrite them.

Other observations about the output later or more likely tomorrow.

Ingress being harder than egress is  another  issue - but one you wouldn't have (in theory) a year ago as Plusnet used to sort QOS for you. Now it seems they don't on the new network - maybe they will fix it, or maybe you still have it and all your issues are egress... I don't know.

Anyone know what the status/plans of Plunet QOS are?

[#2]

Been a bit busy, but noticed tonight that Plusnet QOS is back on my line.

It's different marking to what it used to be but if they get it working for your line (and games) then you shouldn't need to sacrifice any download.

There's still the upstream  issue - if you are up for a bit of command line experimentation there should be a way to work around, unless the router "notices" and resets to its own settings.

[#3]

thanks for very detailed reply, tbh half of it sounds like black magic still

with the new router, setup of bandwidth control needs to be pretty much the same, with around 15000 as max download to keep things under control. quality of wifi is generally better with better hardware of d9, but I haven't noticed much of an improvement when it comes to gaming - it's good now, but still get spikes etc.

how could I tell if my line has plusnet qos applied? (I am actually paying for pro addon hoping to have it better)

also out of pure curiosity I was thinking about a wee project with rebuilding network to look like that;

archer d5 in bridge to work as modem ->

old laptop with pfsense on it to work as router ->

archer d9 as wired and wireless access point

does that have any chance to work better that just a d9?

thanks

[#4]

The Pro Add-on is probably making it worse, because it increases the priority of various different activities to be the same as gaming.

[#5]

As ejs says - drop the pro add on. After that you should be able to work out if downstream QOS is working OK for you.

I know nothing about BSD QOS, so can't say whether that way would be worth the extra hassle, or if you can account for ATM with it (I guess you can, but don't know)

One other thing I notice about your current set up is that the downstream UDP class seems to be getting some bulk traffic (class htb 52:1 on ifb 1)

Sent 13961630732 bytes 12610815 pkt (dropped 1524 .. If there were only game packets going through that then I would expect no drops and a smaller average packet size, though that I guess depends on what games you play - 64 player Battlefield servers do send big packets. If  turning off pro doesn't work then  you need to see about further classification rules to make sure only game traffic gets top prio or maybe raise min rate, but I don't know what games would need that much(3200Kbit). It depends on what is (possibly) causing UDP bulk - IIRC utorrent "invented" some protocol to send bulk over UDP. After saying that, the upstream UDP class doesn't show any sign of bulk. hmm.